Neural Attention Models for Sequence Classification: Analysis and Application to Key Term Extraction and Dialogue Act Detection

Shen, Sheng-syun; Lee, Hung-yi

doi:10.48550/arxiv.1604.00077

Cited by 15 publications

(3 citation statements)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Instead of producing heatmaps [28] or highlighting the words with different colours [29], we directly enlarge and darken the words in the visualization proportional to the attention weights to try reduce the cognitive load of the reader.…”

Section: Analysis Of How Human Interpret Attention From Supervised De...mentioning

confidence: 99%

An empirical user-study of text-based nonverbal annotation systems for human-human conversations

Kim¹,

Yacef²

2021

Preprint

View full text Add to dashboard Cite

With the substantial increase in the number of online human-human conversations and the usefulness of multimodal transcripts, there is a rising need for automated multimodal transcription systems to help us better understand the conversations. In this paper, we evaluated three methods to perform multimodal transcription. They were (1)Jefferson -an existing manual system used widely by the linguistics community, (2) MONAH -a system that aimed to make multimodal transcripts accessible and automated, (3) MONAH+ -a system that builds on MONAH that visualizes machine attention. Based on 104 participants responses, we found that (1) all text-based methods significantly reduced the amount of information for the human users, (2) MONAH was found to be more usable than Jefferson, (3) Jefferson's relative strength was in chronemics (pace / delay) and paralinguistics (pitch / volume) annotations, whilst MONAH's relative strength was in kinesics (body language) annotations, (4) enlarging words' font-size based on machine attention was confusing human users as loudness. These results pose considerations for researchers designing a multimodal annotation system for the masses who would like a fully-automated or human-augmented conversational analysis system.

show abstract

Section: Analysis Of How Human Interpret Attention From Supervised De...mentioning

confidence: 99%

An empirical user-study of text-based nonverbal annotation systems for human-human conversations

Kim¹,

Yacef²

2021

Preprint

View full text Add to dashboard Cite

show abstract

“…This kind of attention mechanism has been integrated with LSTMs in natural language processing (NLP). Shen and Lee [19] applied an attention-mechanism LSTM for key term extraction and dialogue act detection. They showed that the attention mechanism enables the LSTM to ignore noisy or irrelevant passages in long input sequences, as well as to identify important portions of a given sequence to improve sequence labeling accuracy.…”

Section: Attention Mechanismsmentioning

confidence: 99%

“…We compare the performance of the proposed HA-TCN model to that of the classical SVM operating on the handcrafted features described above, and to those of recurrent architectures based on LSTMs and Bidirectional LSTMs (Bi-LSTM), and of recurrent architectures with attention mechanisms like attention-based LSTMs [19] and attentionbased Bi-LSTMs [26], as well as that of the traditional TCN [3]. For deep learning models, each raw handgrip time series is truncated or padded to a total length of 750 time steps, and its amplitude normalized between [0, 1].…”

Section: Dataset and Experimental Setupmentioning

confidence: 99%

Medical Time Series Classification with Hierarchical Attention-based Temporal Convolutional Networks: A Case Study of Myotonic Dystrophy Diagnosis

Lin,

Xu,

et al. 2019

Preprint

View full text Add to dashboard Cite

Myotonia, which refers to delayed muscle relaxation after contraction, is the main symptom of myotonic dystrophy patients. We propose a hierarchical attention-based temporal convolutional network (HA-TCN) for myotonic dystrohpy diagnosis from handgrip time series data, and introduce mechanisms that enable model explainability. We compare the performance of the HA-TCN model against that of benchmark TCN models, LSTM models with and without attention mechanisms, and SVM approaches with handcrafted features. In terms of classification accuracy and F1 score, we found all deep learning models have similar levels of performance, and they all outperform SVM. Further, the HA-TCN model outperforms its TCN counterpart with regards to computational efficiency regardless of network depth, and in terms of performance particularly when the number of hidden layers is small. Lastly, HA-TCN models can consistently identify relevant time series segments in the relaxation phase of the handgrip time series, and exhibit increased robustness to noise when compared to attention-based LSTM models.

show abstract

Conversation Extraction over Short Text Stream with Hierarchical LSTM Neural Networks

Jin

Nie

2019

2019 International Conference on Intelligent Computing, Automation and Systems (ICICAS)

View full text Add to dashboard Cite

Neural Attention Models for Sequence Classification: Analysis and Application to Key Term Extraction and Dialogue Act Detection

Cited by 15 publications

References 14 publications

An empirical user-study of text-based nonverbal annotation systems for human-human conversations

An empirical user-study of text-based nonverbal annotation systems for human-human conversations

Medical Time Series Classification with Hierarchical Attention-based Temporal Convolutional Networks: A Case Study of Myotonic Dystrophy Diagnosis

Conversation Extraction over Short Text Stream with Hierarchical LSTM Neural Networks

Contact Info

Product

Resources

About